FIELD NOTE

Codex CLI 0.159.0 Adds Instant Interrupt and TUI Fixes

Codex CLI 0.159.0 adds opt-in steering during active work, clearer terminal sessions, richer Mermaid diagrams, app-server history pagination and security fixes.

Codex CLI 0.159.0 adds opt-in instant_interrupt, letting new input steer Codex during a model response or a long-running code-mode call. Released September 29, 2026, it also gives new sessions a more compact terminal interface, improves warning and transcript review, expands native Mermaid flowcharts, adds anchored app-server history pagination, and fixes several Windows, macOS, sign-in and filesystem problems. Install the exact release with npm install -g @openai/codex@0.159.0. Read the official ChatGPT and Codex changelog.

The interruption feature is opt-in, so installing 0.159.0 does not by itself prove that active work will yield to a new message. The release notes do not publish the toggle syntax in this entry. Teams should verify the setting in the client they actually run, then test it on a disposable response and a harmless long code-mode call before depending on it for recovery.

Codex CLI 0.159.0 instant interrupt and session flow

instant_interrupt changes a common long-task interaction: instead of waiting for a response or code-mode call to finish, a user can send new input that steers the active work. The useful case is correction, not cancellation for its own sake. If Codex is working from a wrong assumption, a prompt can supply the missing constraint while the turn is still active.

The changelog limits the claim to model responses and long-running code-mode calls. It does not say that every external process, approval dialog or remote job becomes interruptible. A focused check should record what was running, when the new input arrived, whether the active work changed direction, and what final state remained after the intervention.

New sessions also receive a compact welcome screen and consistent headers. Occasional tips can appear during and after turns. These are presentation changes rather than new model capabilities, but they reduce the amount of terminal space spent on startup information and make session state easier to scan.

Warnings, transcript selection and visual output

The warnings viewer now removes reviewed warnings from the viewer when it closes. Press k on a warning that should remain available for later. That creates a deliberate split between acknowledged warnings and retained follow-up items; closing the viewer is no longer the same as keeping its full list untouched.

During the decision about whether to implement a plan, users can scroll the transcript. This makes earlier requirements and evidence available at the point where the choice is made. Copying transcript selections also preserves Markdown tables, formatting and significant whitespace, and more terminals automatically copy selected text. Test a table and an indented block before using copied output as review evidence.

Native Mermaid rendering now supports more flowchart edges, labels and node groups. The previous Codex CLI 0.158.0 release fixed quoted labels and ampersands and explained unsupported fallbacks. Version 0.159.0 extends the supported flowchart surface rather than announcing a new diagram format. A small diagram with labeled edges and a grouped set of nodes is a better upgrade test than a large production chart.

App-server history and blank-session continuity

App-server clients can paginate thread history from a specific item. An integration can anchor a history request to an item instead of repeatedly treating the thread as one undifferentiated list. Client developers should test the anchor at the beginning, middle and current end of a thread, then confirm that no item is skipped or duplicated as pages are combined.

Blank-session handling also improves. Draft text survives when a user switches tasks before the first turn, and a thread can be archived and still appear in listings before it has a populated preview. Those fixes matter to clients that create a thread before the first submitted message. They do not change the meaning of an archived thread or guarantee that a separate client preserves an unsent draft unless it uses the updated behavior.

The earlier Codex CLI 0.157.0 release centered on GPT-6 Sol and Luna, Bedrock support, background sessions, imports and thread forks. Version 0.159.0 does not reannounce those model or provider changes; its thread work is narrower, covering history pagination and empty-thread continuity.

Security and platform fixes in 0.159.0

On Windows, Codex avoids stray console windows when launching MCP servers, code-mode hosts and piped commands. Restrictive launchers can fall back to embedded mode. The fix targets process startup behavior; it does not relax the permissions or approval policy applied to the launched work.

On macOS, the release fixes TLS access in network-enabled sandboxes and remote environments that require proxy access. Local ChatGPT sign-in also opens the browser more reliably, and onboarding offers a shortcut for copying the login link. The GPT-6 access guide can help separate client sign-in from plan, workspace and model availability when diagnosing an access problem.

Filesystem boundaries receive two specific fixes. Approved commands retain explicit filesystem denials instead of losing them during approval preparation, and .aws directories are protected by default when they sit under writable roots. The first preserves a stated exception; the second prevents a broad writable root from silently making AWS configuration writable. Neither change is a reason to grant a broader root than the task needs.

Removed prompt suggestions and bundled plugin creator

Codex CLI 0.159.0 removes automatic follow-up prompt suggestions and the tui.prompt_suggestions setting. A configuration that still names that setting should be treated as stale after the upgrade rather than as proof that suggestions remain enabled. The release also removes the bundled plugin-creator skill. It does not announce that plugin creation itself has been removed, but workflows that depended on that bundled skill should identify a separately available tool or documented process before assuming the old resource is still present.

These removals are the main migration checks because they can invalidate a saved preference or a built-in workflow even when the CLI starts normally. Preserve any local prompt or plugin-development material that the team owns, but do not copy an obsolete bundled skill into a new release without reviewing its current source and compatibility.

What to test after upgrading

Confirm codex --version reports 0.159.0, then test only the paths the team uses. If instant_interrupt is enabled, steer one harmless model response and one disposable code-mode call. Review a warning, keep another with k, scroll the transcript at the plan decision, copy a Markdown table, and render a flowchart with labeled edges and grouped nodes. App-server clients should paginate from a known item and reconcile page boundaries.

On the active operating system, reproduce the relevant launch, proxy, TLS or local sign-in case. Run an approved command with an explicit filesystem denial and verify that the denial survives; where a writable root contains .aws, confirm that the directory remains protected. Finally, check configuration and skill-dependent workflows for the two removals. Codex CLI 0.159.0 is an interaction and reliability release: its material changes are faster steering, clearer terminal review, more precise history access and stronger preservation of existing security boundaries.